Medical intuition is partially subjective to the expertise and experience of medical practitioners. With AI, the same model could be used by different practitioners, which makes the diagnosis more reliable. Furthermore, with biomedical instrumentation, many physically unobservable features could be integrated into AI models. Once optimized and automated, these models and systems are way better than medical intuition. However, the drawback is that data analysis and model validations should be approved first before they get implemented. Therefore medical intuition will continue to be a necessity for some time.

Hi Alae Ammour

Hope these articles may help you:

https://www.advisory.com/daily-briefing/2018/08/03/artificial-intelligence

https://towardsdatascience.com/artificial-intelligence-deep-learning-for-medical-diagnosis-9561f7a4e5f

https://www.healthcarefinancenews.com/news/artificial-intelligence-cant-replace-doctors-gut-instincts-mit-study-says

Regards

I agree with you Alae Ammour, there is medical intuition in good medical diagnoses.

However I have also unfortunately observed the case where a doctor "missed it completely", and prescribed an aspirin to someone who was hospitalised the next day. The doctor had missed symptoms repeated six times in a neurology textbook. The patient survived because additional medical brains could be triggered.

Let us be careful when the medical intuition says "aspirin" to almost everything and "paracetamol" to the rest...

The problem behind this is science and its application.

Science excellence with no application experience is like a software programme doing infinite loops: it goes nowhere.

Application experience with science missed leads to the lack of diagnosis above.

Now what can AI bring?

First the above case would not have been missed: a doctor assisted by AI may be given more choices, or reminders. It's not necessarily good: too much information can be overwhelming.

Finding the right balance is a question of solution engineering: balancing practical medical expertise and computational/information capabilities.

Hi Alae

As long as the data is useful, AI will always be better than an individual doctor. With the correct large sets of data and deep learning networks, individuals will not be able to compete. As I tell my students - imagine going into a room with 100,000 doctors who have your whole family's medical records and can consult with one another. You describe your symptoms (taken by a nurse who feeds it into a database) and the neural network will sift through the data of millions of patients who have similar symptoms. The network can then compare the treatments and outcomes in less than a second. The chances of being diagnosed the wrong illness and more importantly, being prescribed the wrong medicines, will be vastly reduced. I know from personal experience that GPs under huge amounts of pressure get some diagnoses incorrect. I still don't know why they can't admit - "this is not my specialist area - I will have to send you to a consultant." As you can probably tell - I'm hugely impressed with what AI can do. I've just recently discovered that it can even predict chaotic time series!!

Medical intuition is partially subjective to the expertise and experience of medical practitioners. With AI, the same model could be used by different practitioners, which makes the diagnosis more reliable. Furthermore, with biomedical instrumentation, many physically unobservable features could be integrated into AI models. Once optimized and automated, these models and systems are way better than medical intuition. However, the drawback is that data analysis and model validations should be approved first before they get implemented. Therefore medical intuition will continue to be a necessity for some time.

Artificial intelligence guides medical intuition by providing the necessary information (which we would not have thought of) for a diagnosis or scientific conclusions, but AI cannot (for the moment) completely replace the 'medical intuition, they remain complementary.

Both are complementary. AI needs training in terms of the dataset and computational rigorousity whereas any helpful and logical and genuine intuition usually depends upon expertise. Each has its own merits and pitfalls.

You raise a good relevant question in this emerging era of AI and Soft Computing.

Yes, AI is extensively used in the field of medical diagnosis to assist the doctor by virtue of which the errors in diagnosis can be minimized to some extend.

But, AI can not replace the doctor by producing AI Doctor or Machine Intelligent Doctor. Finally, I suggest every doctor need to use AI, Medical expert system to diagnose critical cases for eliminating error in diagnosis.

We can say because it depends on which criteria you want to use to compare different areas but the collaboration of using both of them help out in discovering new solutions.

Machines have some advantages. They can regard very large amount of information about very much parameters. If we have a good model, i think we can get very good predictions. The challenges are both: information and model.

In my opinion, unambiguous, objective resolution of this issue is difficult. Over time and the development of Industry 4.0 technology, including the improvement of artificial intelligence technology and its applications in medicine, an unambiguous, objective answer to this question may be even more difficult. I believe that both medical intuition and the technologies used are important. Therefore, technology should support the work of doctors, nurses and other medical personnel. Technology should be helpful, so technology can support the use of knowledge and also intuition. Technology, including artificial intelligence, can be used as an additional tool to confirm specific diagnosed data, information and the mentioned intuition. The key issue is only to decide what in a given situation is a priority goal to save lives and improve human health. Then, technology should help meet certain priorities. Technology, including artificial intelligence, should be only an instrument that helps to achieve specific goals, which in medicine are saving lives and improving human health.

Greetings,

Dariusz Prokopowicz

I agree with the view that the best option is a combination of both. AI is only a means for accessing information from a database. The database will have errors in diagnosis, it will have geographic and ethnic bias and it will span different standards of medical thinking through time and between regions. AI is good at pattern recognition when trained against appropriately large and high-quality datasets, but its response is only as good as the bespoke interpretation process developed by programmers. If we don’t have doctors in the loop we cannot avoid systematic AI errors. In addition, it is the diversity of human experience and perspectives that drives innovation, which can never be achieved by so-called AI. Therefore, the vast diversity of human knowledge must be passed down through teaching so that medicine (and other professional fields) can continue to advance.

Taken as a whole, in principle yes, in current situation no.

I would like to elaborate on my answer and say that a doctor should be part of any use of AI for medical diagnosis for the foreseeable future. This is my view based on the limitations of our current applied scientific understanding of intelligence. The mathematics of artificial neural networks stems back to the 1940s (e.g. McCulloch and Pitts, 1943). There has been little advancement in understanding the physiology of intelligence since then. Some relevant advancements in the psychology are discussed in the book ‘Thinking Fast and Slow’ (Daniel Kahneman, 2011) and help us conceptualise AI’s limitations. For example, any process a person can do without conscious thought can be described as ‘thinking fast’ and can (probably) be automated using pattern recognition algorithms. However, the process of thinking slow, which solves novel problems and manages the thinking fast process, has no computational analogy. The endocrine system is bound to play a role in animals as this changes motivational state for any situation. Currently, AI requires programmers to define a bespoke prescribed management process for acquiring sensory inputs and generating output actions for all specified situations. AI cannot determine a response for a situation it has not been trained to encounter and displays errors in intuition, just like humans do, when trained against poor quality data. Thus, a doctor should always be in the loop. Condensing human data into an unmotivated, unquestioning, unanswerable pattern recognition algorithm is a poor substitute for professional experience. However, commercial interests and short-term health system economic calculations may decide otherwise, which will be detrimental to medicine assuming this leads to a loss of opportunity to train doctors. Furthermore, AI alone will inevitably provide a poor service to patients with outlying conditions, or pertinent personal circumstances, compared to having a doctor in the loop to navigate the diagnosis process and discuss treatment options.

May be

These arguments suggest that ideas on medicine, etc, are not merely fixed but absolute. I suggest that medical efficacy remains limited and any knowledge obtained by AI now will be obselete in fifty years.

to some extent.

A.I. is worse than human decision-making ("intuitive" or otherwise) in that we do not know who to blame when the decisions of a robot makes a bad call. You cannot punish a machine, nor can you pin the blame on a programmer or manufacturer when the robot in question is genuinely autonomous.

So, while technology comes into play, the use of A.I. in medical diagnosis presents an ethical, not technological, drawback. In fact, I surmise that this "responsibility gap" (as it is called in the literature) will prove decisive in slowing -- and perhaps barring -- a full reliance on A.I. in health matters.

(For more on this, see my forthcoming piece on "The Mandatory Ontology of Robot Responsibility" in the Cambridge Quarterly of Healthcare Ethics...)

I don't think that the issue of a 'responsibility gap' is universally true. National laws and medical employment structures vary around the world. My view is as a layperson. However, I assume that if payment is provided for a service then the organisation using AI to provide the service is accountable where there are comparable existing laws governing negligence. I don't believe that genuine autonomy is a realistic prospect based on current scientific knowledge, but if that changes and if the service is provided free of charge then legal recourse will be aligned with current laws directed to the relevant public sector organisation or charity. Am I missing something?

If this assumption were true at the end there would be no need for doctors.

It is clear that technology, especially computer technology, is capable of greatly helping diagnostic possibilities. But data deriving from experience, from knowledge mediated by human intuition are absolutely irreplaceable.

Also because I think it is much easier for the patient to have compliance with another human being, rather than with a machine.